Model observer task-based assessment of computed tomography metal artifact reduction using a hip arthroplasty phantom

Area of Science:

• Medical Imaging
• Radiology
• Computational Imaging

Background:

• The FDA established a framework for assessing computed tomography (CT) metal artifact reduction (MAR) algorithms using model observers.
• Previous feasibility studies utilized mathematical phantoms for low-contrast detectability (LCD) tasks.

Purpose of the Study:

• To evaluate the feasibility of the FDA's model observer framework for MAR performance assessment in a physical phantom.
• To compare model observer results with human observer performance in detecting lesions in the presence of metal artifacts.

Main Methods:

• A physical hip arthroplasty phantom with simulated lesions was scanned using standard and reduced-dose CT protocols.
• Channelized Hotelling observers (CHO) with Laguerre-Gauss channels were employed for lesion detectability (d') assessment.
• Results were compared to human observer data using Spearman's correlation.

Main Results:

• CHO using Laguerre-Gauss channels and image masking/thresholding provided sensitive MAR performance assessment.
• Iterative MAR (iMAR) significantly improved lesion detectability (d') compared to filtered back projection (FBP) across all dose levels.
• No significant difference in d' was found between full-dose and half-dose protocols for either iMAR or FBP.

Conclusions:

• Model observer-based LCD assessment of MAR performance is feasible using physical phantoms.
• The CHO method demonstrated strong correlation with human observer detection rates and confidence scores.
• This objective framework supports the evaluation of MAR algorithms in realistic imaging scenarios.